Dessert AI — API Playground

Compose the request

Pretend you're a brand's Claude OS calling our API. Pick the variables, then Preview or Run.

1Mode

Existing brand

Bring-Your-Own brand

Client Products (optional — leave empty to let Dessert pick)

Pick a client first

Multi-product mode (NEW — combine 2+ products in one ad)

For Comparison / Bundle: either pick at least 2 products in the chips above, OR leave empty and Dessert auto-picks N distinct products. Picking just 1 product = same product in every slot. Variant: uses Additional Hero Images of a single product (or repeats it).

Pin specific templates by record id (optional — bypasses Active/Alicia filter)

Comma-separated Templates record ids. Use this to test draft / pending-approval templates (e.g. multi-product TPL_517+).

2Output

Format

Aspect ratio

Count (even, min 2)

In-situ render (product placed in lifestyle scene via Gemini) Skip Luma animation (static-only — no $ Luma spend during iteration)

3Creative direction

Style categories — production Animation Prompts

Select any. Pinned via featured_styles override — narrows the eligible Style pool before the orchestrator's scoring runs.

Or natural-language style search

Used only if no Animation Prompts are selected above. Hits the Vertex embedding index.

Copy angle

Brand tone override

4Callbacks

Webhook URL (optional)

When the delivery finishes we POST the rendered URLs here. Otherwise poll /v1/deliveries/<id>.

Live runs hit the production orchestrator. A new Delivery row is created in Airtable and tagged Source = API Playground. Use sparingly on real clients.

Response

Compose on the left, then click Preview to see the request body or Run to fire it.

Request JSON

curl

Full server response

// Compose on the left, then Preview / Run.